[CI] use new docker with uv binary to fix shim/uv didn't exist (#1724)

* [CI] use new docker without pip uv

* fix ipex error

Author: CSY-ModelCloud

.github/workflows/unit_tests.yml

@@ -205,7 +205,7 @@ jobs:
       - list-test-files
     if: needs.list-test-files.outputs.torch-files != '[]' || needs.list-test-files.outputs.transformers-files != '[]'
     container:
-      image: ${{ needs.check-vm.outputs.ip }}:5000/nvidia/cuda:${{ needs.check-vm.outputs.cuda_version }}-ubuntu22.04
+      image: ${{ needs.check-vm.outputs.ip }}:5000/nvidia/cuda:${{ needs.check-vm.outputs.cuda_version }}-ubuntu22.04_0822
       options: --device /dev/dri --ipc=host --runtime=nvidia --gpus all
       volumes:
         - /dev/dri/by-path:/dev/dri/by-path
@@ -327,7 +327,7 @@ jobs:
       - check-vm
     runs-on: [ self-hosted, xeon5 ]
     container:
-      image: ${{ needs.check-vm.outputs.ip }}:5000/nvidia/cuda:${{ needs.check-vm.outputs.cuda_version }}-ubuntu22.04
+      image: ${{ needs.check-vm.outputs.ip }}:5000/nvidia/cuda:${{ needs.check-vm.outputs.cuda_version }}-ubuntu22.04_0822
       options: --device /dev/dri --ipc=host --runtime=nvidia --gpus all
       volumes:
         - /dev/dri/by-path:/dev/dri/by-path

tests/test_packable.py

@@ -8,7 +8,7 @@
 from gptqmodel.nn_modules.qlinear.exllama import ExllamaQuantLinear  # noqa: E402
 from gptqmodel.nn_modules.qlinear.exllama_eora import ExllamaEoraQuantLinear
 from gptqmodel.nn_modules.qlinear.exllamav2 import ExllamaV2QuantLinear  # noqa: E402
-from gptqmodel.nn_modules.qlinear.ipex import IPEXQuantLinear  # noqa: E402
+from gptqmodel.nn_modules.qlinear.ipex import IPEXQuantLinear, HAS_IPEX  # noqa: E402
 from gptqmodel.nn_modules.qlinear.marlin import MarlinQuantLinear  # noqa: E402
 from gptqmodel.nn_modules.qlinear.torch import TorchQuantLinear  # noqa: E402
 from gptqmodel.nn_modules.qlinear.tritonv2 import TritonV2QuantLinear  # noqa: E402
@@ -59,6 +59,8 @@ def setUpClass(cls):
         ]
     )
     def test_post_init(self, backend: BACKEND, equal: Dict[str, bool]):
+        if backend == BACKEND.IPEX and not HAS_IPEX:
+            self.skipTest("IPEX is not available")
         model = GPTQModel.load(self.model_id, backend=backend, device_map="auto")
         model = convert_gptq_v2_to_v1_format(model, model.quantize_config, self.QLINEAR_DICT[backend])